In view of the problem of detecting and tracking maneuvering small targets at low signal-to-noise, a track-before-detect algorithm based on sequential Monte Carlo probability hypothesis density filtering for Jump-Markov systems (JMS-SMC-PHD) is presented. Under the condition of an unknown number of maneuvering targets and unknown models, the algorithm achieves track-before-detect of small maneuvering targets by using measurement data from infrared sensors directly, adding a variable that denotes the dynamics model of the target, and using a Markov model probability transfer matrix combined with an SMC-PHD filter. Simulation results show that the proposed method can effectively implement target detection and tracking performance.